Change from nvidia-docker to plain docker

[helmuthb]

When using the latest version from NVIDIA Docker, the command nvidia-docker is deprecated. Instead, one is supposed to use regular docker with the --gpus option.

This also adds libgl1-mesa-glx in the docker container for fixing issue #9.

---

This change is 

[Congdinh1801]

@helmuthb I was trying your changes. Got this error when I run "make docker-run-test-sample"

Done with data preparation and model configuration.
Done with model inference.
Process and visualizing the outputs...
Saved semantic segmentation visualization in semantic_segmentation_result.jpg
Traceback (most recent call last):
  File "scripts/demo.py", line 100, in <module>
    demo()
  File "scripts/demo.py", line 94, in demo
    det_vis = visualize_detection_image(instance_detection[0], input_image, cityscapes_instance_label_name)
  File "/workspace/panoptic/realtime_panoptic/utils/visualization.py", line 263, in visualize_detection_image
    cv2.putText(visualized_image, s, (x, y), cv2.FONT_HERSHEY_SIMPLEX, .5, (255, 255, 255), 1)
cv2.error: OpenCV(4.5.3) :-1: error: (-5:Bad argument) in function 'putText'
> Overload resolution failed:
>  - Can't parse 'org'. Sequence item with index 0 has a wrong type
>  - Can't parse 'org'. Sequence item with index 0 has a wrong type

make: *** [Makefile:34: docker-run-test-sample] Error 1

[jieli352]

Hi @Congdinh1801, That's probably a problem due to OpenCV version. Would you try adding int() wrapper around the coordinates in this line? E.g. cv2.putText(visualized_image, s, (int(x), int(y)), cv2.FONT_HERSHEY_SIMPLEX, .5, (255, 255, 255), 1)

[Congdinh1801]

@jieli352 it works now thanks a lot!